Board of Governors of the Federal Reserve System

International Finance Discussion Papers
Number 907
October 2007




A Residual-Based Cointegration Test for Near Unit Root Variables

Erik Hjalmarsson and Pär Österholm











NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate
discussion and critical comment. References in publications to International Finance Discussion Papers
(other than an acknowledgment that the writer has had access to unpublished material) should be cleared
with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/.
A Residual-Based Cointegration Test for Near Unit Root
Variables
+
Erik Hjalmarsson
y
Pär Österholm
z
October 9, 2007
Abstract
Methods of inference based on a unit root assumption in the data are typically not robust to
even small deviations from this assumption. In this paper, we propose robust procedures for a
residual-based test of cointegration when the data are generated by a near unit root process. A
Bonferroni method is used to address the uncertainty regarding the exact degree of persistence in
the process. We thus provide a method for valid inference in multivariate near unit root processes
where standard cointegration tests may be subject to substantial size distortions and standard OLS
inference may lead to spurious results. Empirical illustrations are given by: (i) a re-examination
of the Fisher hypothesis, and (ii) a test of the validity of the cointegrating relationship between
aggregate consumption, asset holdings, and labor income, which has attracted a great deal of
attention in the recent …nance literature.
JEL classi…cation: C12, C22.
Keywords: Bonferroni test; Cointegration; Near unit roots.

We have bene…tted from comments by Meredith Beechey, David Bowman, Mike McCracken, Chris Erceg, Dale
Henderson, Lennart Hjalmarsson, Randi Hjalmarsson, George Korniotos, Rolf Larsson, Andy Levin, Johan Lyhagen,
John Rogers, Jonathan Wright, and seminar participants at the Federal Reserve Board. The views in this paper are
solely the responsibility of the authors and should not be interpreted as re‡ecting the views of the Board of Governors
of the Federal Reserve System or of any other person associated with the Federal Reserve System. Österholm gratefully
acknowledges …nancial support from Jan Wallander’s and Tom Hedelius’ Foundation.
y
Division of International Finance, Federal Reserve Board, Mail Stop 20, Washington, DC 20551, USA. email:
erik.hjalmarsson@frb.gov Phone: +1 202 452 2426
z
Department of Economics, Uppsala University, Box 513, 751 20 Uppsala, Sweden. email: par.osterholm@nek.uu.se
Phone: +1 202 378 4135
1 Introduction
Cointegration tests have been among the most important and in‡uential tools in empirical economics
since their introduction over two decades ago. In essence, cointegration tests attempt to identify com-
mon driving factors in stochastically trending data, thus identifying long-run equilibrium relationships
between economic variables. The most common cointegration tests are based on the assumption that
the individual variables are unit root processes. The unit root assumption, however, is often hard to
fully justify for actual economic data. In …nite samples, many economic variables appear highly, but
not totally, persistent; that is, the largest autoregressive root is close to, but not necessarily equal
to, unity. Unfortunately, inferencial procedures designed for unit root data tend not to be robust to
even small deviations from the unit root assumption. For instance, Elliot (1998) shows that large size
distortions can occur when performing inference on the cointegrating vector in a system where the
individual variables follow near unit root processes rather than pure unit root processes.
Unit root tests go some way toward alleviating the uncertainty regarding the persistence in a given
time series but do not provide a de…nitive answer. Since unit root tests have low power against local
alternatives, a failure to reject the null hypothesis of a unit root does not rule out the possibility of a
root slightly di¤erent from unity. On the other hand, rejecting the null of a unit root does not rule out
that the process is still fairly persistent and leaves open the possibility of spurious regressions. It is
thus far from obvious how to deal with a multivariate near unit root process: Standard cointegration
tests will not be valid under deviations from the pure unit root assumption and the possibility of
spurious regressions invalidates standard OLS inference.
1
The aim of this paper is to design a test of cointegration that is robust to deviations from the pure
unit root assumption. In particular, we extend the standard framework to the case where the original
data possess autoregressive roots that are local-to-unity, rather than identically equal to unity. The
methods developed here are useful from two di¤erent perspectives. First, they provide a robustness
check to standard cointegration tests in the typical situation where it is not known with certainty that
there is an exact unit root in the data. Second, and just as importantly, the test procedures in this
paper allow for valid inference in the case when the data is likely not a pure unit root process, but still
highly persistent.
1
In most cointegration studies, the regressors are endogenous, in which case OLS inference would be further compli-
cated and invalid even in the strictly stationary case. Stock (1997) provides a detailed discussion on many of the issues
that arise in inference with near unit root variables.
1
While there is a large literature on cointegrating regressions with near unit root regressors, the
focus has been on inference on the slope parameter in these regressions, rather than actual tests of
cointegration; see, for example, Cavanagh et al. (1995), Elliot (1998), Campbell and Yogo (2006) and
Jansson and Moreira (2006). Typically, the models in this literature have been speci…ed such that under
the null hypothesis of a zero slope coe¢cient, the dependent variable is a stationary process. Tests
on the slope coe¢cient therefore become joint tests of cointegration as well, and the issue of spurious
regressions never occurs. Although this is a useful speci…cation, for instance, in tests of stock-return
predictability which motivated much of this literature, it is less convenient in most typical economic
applications where both dependent and independent variables are near-integrated. The closest related
literature to the current paper is the work on stationarity tests (Leybourne and McCabe, 1993, and
Shin, 1994) and the work by Wright (2000). In particular, Wright (2000) develops a joint test of a
speci…c hypothesis regarding the cointegrating vector and a test of the null hypothesis of cointegration
that is robust to deviations from the pure unit root framework.
We focus on a residual-based test of cointegration. Following the work of Phillips and Ouliaris
(1990), we extend the asymptotic results for a residual-based test to the case of near-integrated
processes. Unlike the pure unit root case, the asymptotic distribution of the test statistic now depends
on an unknown nuisance parameter; the local-to-unity root. Since this parameter is not consistently
estimable, feasible tests cannot be directly constructed from the asymptotic distribution. Instead,
we propose to replace the unknown parameter value for the local-to-unity root with a conservative
estimate.
In order to understand the intuition behind our procedure, it is useful to consider the potential errors
when applying a standard, pure unit root case, cointegration test to a set of near unit root variables. A
residual-based cointegration test evaluates whether the residuals from the empirical regression contain
a unit root. Now, if the original data are in fact near-integrated, with a root less than unity, the test
will over-reject since the residuals will not contain a unit root even if there is no cointegration. But, by
instead using critical values based on a conservative estimate of the local-to-unity root in the original
data, a valid test is obtained. Intuitively, if one views a residual-based test of cointegration as a test
of whether there is less persistence in the residuals than in the original data, then this test is only
valid if the persistence of the original data is not overstated.
2
In a spirit similar to the Bonferroni
2
Although, perhaps, less obvious, the same also holds true for non-residual-based tests, such as those of Johansen
(1988,1991); see Hjalmarsson and Österholm (2007).
2
methods proposed by Cavanagh et al. (1995), we show how an appropriately conservative estimate of
the local-to-unity root is obtained.
The rest of the paper is organized as follows. Section 2 outlines the modelling assumptions and the
theoretical results. Section 3 describes the Bonferroni methods. In Section 4, the proposed procedure
is evaluated using Monte Carlo simulations. We show that once the conservative estimate for the
local-to-unity parameter is chosen appropriately, the resulting test has both good size and power
properties. This is in contrast to standard cointegration tests, based on the unit root assumption,
which are shown to severely over-reject as the data generating process deviates from a pure unit root
setup. As an illustration of the method, two empirical applications are considered in Section 5. First,
we re-examine the Fisher hypothesis and show that using the robust methods proposed in this paper,
one can no longer …nd signi…cant support for a long-run equilibrium relationship between nominal
interest rates and in‡ation; using standard unit root based cointegration tests on the other hand, the
null hypothesis of no cointegration is rejected. In a second illustration, we consider the robustness
of the long-run relationship between aggregate consumption, asset holdings, and labor income, which
was initially studied by Lettau and Ludvigson (2001) and has since received a great deal of attention
in the …nance literature. We …nd that after controlling for the unknown persistence in the variables,
there is still strong evidence of cointegration between the three variables. Section 6 concludes and the
Appendix contains tables of critical values for the test statistic.
2 Theoretical framework
2.1 Model and assumptions
Let ¦.
t
¦
1
0
be an :÷vector of nearly integrated processes, such that the data generating process satis…es
.
t
= ¹.
t1
+n
t
(1)
where ¹ = 1 + C/ T is an :: matrix with ¹ = diao (a
1
. .... a
m
) and C = diao (c
1
. ... c
m
), and T is
the sample size. That is, each component process in .
t
is generated as a near unit root process with
individual local-to-unity parameters c
i
, i = 1. .... :. The initial conditions are set at t = 0 and .
0
is
assumed randomly distributed with …nite variance. Although none of the formal results depend upon
it, we will work under the assumption that c
i
_ 0 for all i, which rules out explosive processes. The
3
innovations n
t
satisfy a general linear process.
Assumption 1 1. n
t
= 1(1) c
t
=
¸
1
j=0
1
j
c
tj
.
¸
1
0
, [[1
j
[[ < ·. [1(1)[ = 0.
2. c
t
is iid with mean zero, variance matrix

, and …nite fourth-order moment.
By standard results, e.g. Phillips and Solo (1992), T
1=2
¸
[Tr]
t=1
n
t
=1(r) = 1' (), where 1(r)
is a Brownian motion with covariance matrix = 1(1)

1(1)
0
. Partition .
t
= (n
t
. r
0
t
)
0
such that n
t
is
a scalar and r
t
is an :÷vector (: = :÷1). Let 1(r) =

1
1
(r) . 1
2
(r)
0

0
, = [(.
11
. .
0
21
) . (.
21
.
22
)],
and C = [(c
1
. 0) . (0. C
2
)] be conformable partitions of 1(r), , and C, respectively. We assume that

22
0 and write = 1
0
1. Denote an :÷vector standard Brownian motion as \ (r), and it follows
that 1(r) = 1
0
\ (r). Further, as T ÷ ·, .
t
/

T = J
C
(r) =

r
0
c
(rs)C
d1(:). Partition J
C
conformably with 1 and let J
W
C
(r) =

r
0
c
(rs)C
d\ (:) .
We consider residual-based tests of the null of no cointegration using the regression residuals, ^ ·
t
,
from the following empirical regression:
n
t
=
0
r
t

t
. (2)
2.2 The test statistic
We focus on the traditional Augmented Engle-Granger t÷test (Engle and Granger, 1987) of the null of
no cointegration, which is probably the most commonly used residual-based test of cointegration. Our
analysis could easily be extended to cover the 7

and 7
t
cointegration tests proposed by Phillips and
Ouliaris (1990), but for brevity we restrict ourselves to the Augmented Engle-Granger test (henceforth
denoted AEG test).
The AEG test is de…ned as the t÷statistic for ^ c

from the regression ^ ·
t
= c

^ ·
t1
+
¸
p
i=1
.
i
^ ·
ti
+
n
t
. The below result follows from the results in Phillips and Ouliaris (1990) and the results for near-
integrated processes in Phillips (1987,1988).
Theorem 1 Let the data generating process satisfy equation (1) for some given C = diao (c
1
. .... c
m
),
and let Assumption 1 hold. Suppose that the autoregressive order in the AEG regression satis…es
4
j ÷· as T ÷· such that j = o

T
1=3

. Then, under the null of no cointegration, as T ÷·,
¹1G =c
1

1
0

J
W
12;C

2
1=2
+

1
0
J
W
12;C
d\
12

1
0

J
W
12;C

2

1=2
(3)
where
J
W
12;C
(r) = J
W
1;c1
(r) ÷

1
0
J
W
1;c1
J
W0
2;C2

1
0
J
W
2;C2
J
W0
2;C2
1
J
W
2;C2
(r) (4)
and
\
12
(r) = \
1
(r) ÷

1
0
\
1
\
0
2

1
0
\
2
\
0
2
1
\
2
(r) . (5)
are the L
2
÷projection residuals of J
W
1;c1
and \
1
on the spaces spanned by J
W
2;C2
and \
2
respectively.
Remark 1.1 The limiting distribution of the AEG statistic depends on the unknown parameter C,
but is otherwise free of nuisance parameters. For a given C, the asymptotic distribution can thus easily
be tabulated. The next section describes a feasible implementation of the test when C is unknown.
Remark 1.2 E¤ectively, the AEG test evaluates whether the persistence in the residuals is less than
that predicted under the null hypothesis of no cointegration. However, since the original data is not
necessarily a unit root process, the critical values re‡ect this fact. In the special case of C = 0, the
limiting distribution reduces to the usual one for pure unit root variables.
Remark 1.3 In empirical work, a constant or a constant and a linear trend are typically included
in the empirical regression (2). As in standard cointegration analysis, this will a¤ect the limiting
distribution in a straightforward manner (e.g. Phillips and Ouliaris, 1990) and thus the critical values
used, but will otherwise not alter the analysis.
3 Feasible implementation
For a known C, the above test is trivial to use once critical values for the asymptotic distribution are
obtained. Unfortunately, C is typically not known. We therefore consider a Bonferroni test approach,
which is similar to that used by Cavanagh et al. (1995) and Campbell and Yogo (2006) in their pursuit
of inference in predictive regression with near-integrated variables.
5
Consider con…dence intervals for c
i
, i = 1. .... :, of the shape ¦[c
i
. c
i

m
i=1
with an overall coverage
rate equal to 100 (1 ÷c
1
) percent. Let ¦¯c
i
÷ [c
i
. c
i

m
i=1
be the set of parameter values in this
con…dence region for which the critical value of the asymptotic distribution of the ¹1G statistic is
most conservative, for some given c
2
percent level (e.g. …ve percent). If the ¹1G statistic is evaluated
using this conservative critical value, calculated at the c
2
percent level, the size of the resulting
cointegration test will be less than or equal to c = c
1
+c
2
, by Bonferroni’s inequality.
However, relative to the Cavanagh et al. (1995) and Campbell and Yogo (2006) studies, there is an
additional complication in the current setup. In those papers, there is only one local-to-unity process,
whereas here there are at least two in the simplest case with just one regressor. In the univariate
case, con…dence intervals of the local-to-unity parameter can be obtained by inverting a unit root test
statistic (Stock, 1991). In the :÷dimensional case, a con…dence region for C could be obtained by
inverting individual unit root test statistics in order to obtain con…dence intervals [c
i
. c
i
], i = 1. .... :,
each with coverage rate 1 ÷c
1
´:. The overall con…dence level of ¦[c
i
. c
i

m
i=1
is at least 100 (1 ÷c
1
)
percent, again by Bonferroni’s inequality. Although theoretically sound, such an approach su¤ers from
the practical disadvantage that it would be virtually impossible to tabulate the critical values for
the asymptotic distribution beyond the simple two-dimensional case. We therefore propose a simpler
approach that allows for tabulation of critical values and seems to give up little in robustness.
Intuitively, the AEG test evaluates whether the persistence, or autoregressive root, in the regression
residuals, ·
t
, is less than in the original data, n
t
. As seen in equations (3) and (4), the critical values of
the test depend on both the persistence in the ‘dependent’ variable, n
t
, and the regressors, r
t
, denoted
c
1
and C
2
respectively. However, it seems reasonable to conjecture that the main determinant of the
asymptotic distribution will be c
1
, rather than C
2
. Thus, using
^
C =
^
C
1
= diao (^ c
1
. .... ^ c
1
) for some ^ c
1
,
rather than
~
C = diao (~ c
1
. ~ c
2
. .... ~ c
m
), to form critical values might not cause a large size distortion in
the test. Although this conjecture is di¢cult to evaluate analytically, extensive simulation evidence
supports it. For instance, Figure 1 shows the critical values for the AEG test in the two-dimensional
case with an intercept in the empirical regression. As is evident, the primary changes come from
changing c
1
, whereas the critical values are almost constant across C
2
. Additional evidence supporting
this conclusion is provided by simulations in the following section.
Furthermore, if C
1
= diao (c
1
. .... c
1
) is used to calculate the critical values for the asymptotic
distribution in Theorem 1, the ¹1Gcointegration test will be more conservative as the value of c
1
6
decreases; that is, as c
1
becomes more negative, so do the corresponding critical values, as shown in
Table A3. Only the lower bound on c
1
, say c
1
, is therefore of interest in constructing a conservative
test; for a given con…dence level, such a lower bound can be obtained from a one-sided con…dence
interval for c
1
, [c
1
. +·).
By restricting the attention to the parameter c
1
, and calculating critical values based on C
1
=
diao (c
1
. .... c
1
), it now becomes easy to implement the Bonferroni method. The lower con…dence
bound for c
1
, c
1
, is obtained by inverting a unit root test statistic for the variable n
t
. Based on this
lower bound of c
1
, the test is evaluated using the corresponding critical value for C
1
= diao (c
1
. .... c
1
).
If the lower bound c
1
has con…dence level 1 ÷ c
1
and the AEG test is evaluated at the c
2
level, the
resulting test will have a size no larger than c = c
1
+c
2
.
3
In general, Bonferroni’s inequality is strict, and the size of the test will be less than c. To obtain
a correctly sized test of size ~ c, which is distinct from c = c
1
+ c
2
, we …rst …x c
2
at some level and
then …nd c
1
such that the resulting test has size ~ c. Finding c
1
is e¤ectively a trial and error exercise.
In the simulations below, we let ~ c = c
2
= 0.05 and show that setting c
1
equal to 50 percent will
approximately result in an overall …ve percent test. Thus, by e¤ectively using a median unbiased
estimate of c
1
, an approximately correctly-sized test is obtained. These results are discussed more
extensively in conjunction with the Monte Carlo simulations in the next section.
In terms of practical implementation, we follow Campbell and Yogo (2006) and invert Elliot et
al.’s (1996) DF-GLS unit root test statistic to obtain a lower bound for c
1
. Table A1 provides the
lower 95th, 75th, 50th, 25th, and 5th percent con…dence bounds of c
1
, given a value of the DF-GLS
test statistic.
4
For instance, the lower con…dence bound that corresponds to c
1
= 0.05 is given in the
100(1 ÷c
1
) % = 95% column. Table A2 provides the corresponding bounds when a trend is allowed
for in the DF-GLS regression. Table A3 tabulates the …ve percent critical values for the AEG statistic,
for c
1
= 0 to c
1
= ÷60, assuming that c
1
= c
2
= ... = c
m
; values for one to …ve regressors are provided
for the cases of no intercept, intercept, and intercept and a linear trend in the empirical regression.
Henceforth, we will refer to the cointegration test constructed in the manner above as the Bonferroni
AEG test, with the additional speci…cation of the value of c
1
when necessary. Unless otherwise noted,
we let c
2
= 0.05.
3
Since C
2
is assumed not to play an important role in the distribution of the test-statistic, the only uncertainty
regarding the persistence of the data comes from uncertainty regarding c
1
. The con…dence level of the lower bound C
1
is therefore 1
1
rather than 1 m
1
, as discussed above.
4
Note that, for instance, the two lower con…dence bounds at the 5 percent and 95 percent level provide a two-sided
con…dence interval with con…dence level 90 percent.
7
4 Finite-sample properties
4.1 Size properties
We analyze the …nite-sample properties of the proposed test procedure through a series of Monte Carlo
simulations. Starting with the size properties, it is assumed that the data generating process (DGP)
is given by equation (1), with the innovations n
t
drawn from a multivariate normal distribution such
that 1 [n
t
] = 0 and 1 [n
t
n
0
t
] = 1. The sample size is set to either T = 100 or 500 and the number of
regressors, : = :÷1, is equal to either one or three. The regression
n
t
= c +
0
r
t

t
(6)
is estimated, which is a spurious regression given the above DGP, and the cointegration tests are
applied to the …tted residuals, ^ ·
t
. Each simulated :÷dimensional time-series .
t
is thus partitioned
as .
t
= (n
t
. r
0
t
)
0
, as described previously. When all components in .
t
are ex-ante identical, i.e. have
the same persistence c, the …rst component series is set to n
t
and the remainder to r
t
. When c
i
varies
between each series, we describe explicitly which series are set as n
t
and r
t
. All tests are performed at
the …ve percent signi…cance level and are evaluated using the critical values given in Table A3. The
results are based on 10. 000 repetitions.
In the …rst round of simulations, we let the local-to-unity matrix for .
t
be given by C = diao (c. .... c),
so that all the series have identical persistence. The local-to-unity parameter c varies from 0 to ÷30.
Figure 2 shows the size properties for the traditional ¹1G cointegration test, which by de…nition
is evaluated at c = 0, as a function of the local-to-unity parameter c. The nominal size of the test is
…ve percent, and for c close to zero, the actual rejection rate is also close to …ve percent. However, as
c decreases in value, the test starts over-rejecting and the rejection rates already approach ten percent
for c = ÷5. The rejection rates become even larger and approach one as c becomes even smaller. It
should be stressed that this is not a small-sample bias, but a re‡ection of the inconsistency of the test
when c < 0. Since the autoregressive root of the residual in equation (6) is less than one for c < 0, the
AEG test, evaluated under the assumption of c = 0, will reject the null of a unit root in the residuals
more frequently than its nominal size. For time series that do not necessarily have a unit root, standard
cointegration tests can thus be highly misleading. This raises questions regarding previous studies that
have relied on cointegrating methods, despite having found evidence of stationarity of the included
8
variables; see, for example, Crowder and Ho¤man (1996).
We next consider the size properties of the Bonferroni AEG test using a conservative estimate of
C. As discussed in the previous section, we use C
1
= diao (c
1
. .... c
1
) where c
1
is the lower bound on
the persistence in n
t
. A direct application of the Bonferroni method suggests choosing c
1
such that the
one-sided con…dence interval [c
1
. +·) has con…dence level 100(1 ÷c
1
) percent, and then evaluating
the AEG test-statistic at the c
2
percent level for a total size of c = c
1
+ c
2
percent. In practice,
however, such an approach will deliver extremely conservative tests. For instance, if c
1
= c
2
= 0.05,
the rejection rate for the resulting test is virtually identical to zero in the simulations considered here.
Instead, we follow the approach outlined above and …x c
2
= 0.05 and choose c
1
such that the size
of the overall test is close to …ve percent. In particular, we consider setting c
1
= 0.25. 0.50 and 0.75.
That is, c
1
is chosen as the lower bound in one-sided con…dence intervals with con…dence levels equal to
75, 50, and 25 percent, respectively. To obtain these values for c
1
, the DF-GLS unit root test-statistic
is inverted, using the values in Table A1.
5
Figure 3 shows the results for the Bonferroni AEG test using these di¤erent estimates of C
1
. It
is immediately apparent that for small values of c, the test tends to over-reject when c
1
= 0.75, and
under-reject when c
1
= 0.25. For c
1
= 0.50, the test still tends to under-reject somewhat, except
for small values of c in the case of T = 500 and : = 1, where there is instead a slight over-rejection.
Overall, however, for c
1
= 0.50, the rejection rate is typically between two and …ve percent. One
could achieve rejection rates that are somewhat closer to the nominal size by letting c
1
vary with c
1
in some manner, but at the cost of a substantially more cumbersome procedure. Using a …xed value
of c
1
= 0.50, for all values of c
1
, yields a very simple test to implement. The procedure would simply
be given as:
(i) Obtain the value of the AEG test statistic from a standard implementation of the Engle and
Granger test.
(ii) Calculate the DF-GLS unit root statistic for the n
t
variable and obtain the corresponding value
of c
1
from Table A1 or A2.
(iii) Compare the AEG test statistic to the critical value corresponding to c
1
in Table A3.
5
The number of lags included in the DF-GLS test is chosen using the Schwarz (1978) information critierion, with a
maximum number of two allowed in order to keep the simulation times managable. The same number of lags is also
included in the AEG regression; that is, in ^ vt = ^ v
t1
+
P
p
i=1
'
i
^ v
ti
+wt.
9
It may seem surprising that using, for instance, a lower bound with only a 25 percent con…dence
level, does not result in a larger size distortion. Figure 4 helps shed some light on this puzzle. The
results in the …gure are based on 10. 000 simulations of a univariate local-to-unity process, with local-
to-unity parameter c, iid normal innovations and sample size T = 500. It shows estimates of the
lower bounds of c, with con…dence levels of 25, 50, and 75 percent, using the inversion of the DF-GLS
statistic in Table A1. The panels in Figure 4 show the densities for the lower-bounds estimates for
c = ÷5. ÷10. ÷20. and ÷30. As expected, the bounds estimates at the 75 percent con…dence level are
furthest to the left. However, the densities are far from symmetric, especially for c close to zero; the
density for the 25 percent con…dence bound is also less symmetric than the density for the 75 percent
bound. Thus, although the density is shifted further to the right as the con…dence level decreases,
which leads to estimates of c closer to zero, the shift is not symmetric and the risk of vastly over-
estimating c is not increased dramatically. This explains, to some extent, why the rejection rates in
the cointegration test only increase slowly as the con…dence level of the lower bound is decreased.
In the last set of size simulations, shown in Figure 5, we analyze the properties of the Bonferroni
AEG test when the local-to-unity parameters c
i
, i = 1. .... : are not identical; i.e. when the processes
in .
t
do not have the same persistence. Two di¤erent cases are considered. In the …rst case, there
are two regressors with persistence parameters equal to ÷10 and ÷20. In the second setup, there are
three regressors with persistence parameters 0. ÷10, and ÷20. In both cases, it is assumed that the
persistence in n
t
, c
1
, varies between 0 and ÷30. Thus, in the …rst case, C = diao (c
1
. ÷10. ÷20), and
in the second case C = diao (c
1
. 0. ÷10. ÷20). The same methods as in the case with identical c
i
:
are used and the results for c
1
= 0.25. 0.50. and 0.75, are shown. Overall, the results in Figure 5
are very similar to those in Figure 3. Using c
1
= 0.50 and a nominal size of …ve percent results in
actual rejection rates around three percent. Given the results shown previously in Figure 1, it is not
surprising that the test also performs well when the c
i
: are not identical.
In summary, the proposed procedure for tests of cointegration in data with an unknown C appears
to work well in …nite samples, once the con…dence level of the lower bound is chosen appropriately.
Additional …ne tuning of this con…dence level could be done to bring the actual size even closer to the
nominal size, but at the cost of adding some complexity.
10
4.2 Power properties
We next perform a second Monte Carlo simulation to evaluate the …nite-sample rejection rates under
the alternative of cointegration. The ‘independent’ variable r
t
is still generated according to equation
(1) using iid standard normal innovations. However, the ‘dependent’ variable n
t
, is now generated as
n
t
=
0
r
t

t
. (7)
where ·
t
is an ¹1(1) process with an auto-regressive root j; the innovations to this ¹1 process are
iid standard normal. is set to an :-vector of ones. The same empirical regression, including the
constant, as in the size simulations is estimated, and the Bonferroni AEG test with c
1
= 0.50 is
applied to the estimated residuals ^ ·
t
. The critical values that are used are thus for the case with a
constant in the regression. Two di¤erent sample sizes, T = 100 and 500, and : = 1 and 3 regressors,
are considered. In the case of one regressor, the persistence in r
t
is set equal to either C
2
= ÷2. ÷10.
or ÷20. In the case of three regressors, it is assumed that C
2
= diao (0. ÷10. ÷20).
Figure 6 shows the results in four sub-plots corresponding to the di¤erent combinations of sample
size and number of regressors. The vertical axes of the graphs show the power of the Bonferroni AEG
test plotted against the persistence j in the error term ·
t
. In the case of T = 100, results for j ÷ [0.5. 1]
are shown and for the T = 500, results for j ÷ [0.8. 1] are shown. As is to be expected, power is a
monotone and declining function of the persistence, j. It should be noted that for very large values
of j, we expect the test to have low power; for example, in the bivariate case, a residual that is less
persistent than n
t
cannot be generated by regressing n
t
on r
t
when j 1 +C
2
´T. For most values of
j, however, the test appears to exhibit good power properties and appears su¢ciently powerful that
it would be a useful tool in many empirical applications, including those with relatively small sample
sizes.
5 Empirical illustrations
To illustrate the empirical use of the Bonferroni AEG test, we next consider two applications where
the variables in question are all fairly persistent, but not necessarily pure unit root processes. As a
comparison to the robust methodology proposed in this paper, we will also conduct the traditional
AEG test.
11
5.1 The Fisher hypothesis
It is well known that both nominal interest rates and in‡ation are fairly persistent in most countries.
Accordingly, cointegration techniques have been a popular approach to test the Fisher hypothesis in
more recent years; see, for example, Mishkin (1992), Wallace and Warner (1993), Evans and Lewis
(1995), and Crowder and Ho¤man (1996). However, the assumption made in most of these studies of
exact unit roots in both nominal interest rates and in‡ation can be questioned on both theoretical and
empirical grounds.
6
It is therefore worth re-interesting this issue using the Bonferroni AEG test.
A common formulation of the Fisher hypothesis is that the :-period nominal interest rate (i
m
t
) is
related to the real interest rate (r
m
t
) and in‡ation (¬
m
t
) according to
i
m
t
= 1
t
(r
m
t
) +1
t

m
t
) . (8)
Relying on the commonly made assumption of a constant or mean-reverting real interest rate, an
empirical version of the Fisher hypothesis can be written as
i
m
t
= c +¬
m
t

t
. (9)
where the constant c has the interpretation of the (constant) equilibrium real interest rate, the error
term ·
t
is assumed to be a stationary ARMA process and , in the most traditional interpretation,
should be equal to unity.
7
Monthly data on the short nominal interest rate – given by the three month treasury bill – and
CPI in‡ation from January 1955 to October 2006 in the United States were provided by the Board
of Governors of the Federal Reserve System. Table 1 shows the results from the DF-GLS unit root
test and the KPSS stationarity test, as well as the median unbiased estimate of c, denoted ^ c, and
a 90 percent con…dence interval for c; the estimates and con…dence intervals of c are derived using
the values in Table A1 and linear interpolation.
8
As can be seen, the evidence for a unit root in the
interest rate appears reasonably strong; the DF-GLS test fails to reject the null of a unit root whereas
the KPSS test rejects the null of stationarity. For in‡ation, on the other hand, the evidence is more
6
See, for example, Wu and Zhang (1996), Culver and Papell (1997), and Wu and Chen (2001).
7
Note that in the estimations below, time t in‡ation is given as future in‡ation between t and t + m. This can be
motivated by assuming rational expectations; see, for example, Mishkin (1992).
8
Regarding the speci…cation of deterministic terms in the unit root tests, it should be noted that we test for mean
reversion around a constant level.
12
mixed since the DF-GLS test rejects a unit root but the KPSS test rejects stationarity.
9
Table 1: Unit root tests.
i
t
¬
t
DF-GLS -1.40 -2.54

KPSS 0.53

0.52

^ c -3.40 -12.91
90% CI for c [-9.06, 2.00] [-21.37, -3.46]
Notes: * indicates signi…cance at the …ve percent level.
The cointegration tests are conducted using a signi…cance level of …ve percent. For the Bonferroni
AEG test, based on the simulation results in the previous section, we set c
1
= 0.5; thus ^ c = ÷3.40,
the median unbiased estimate for the nominal interest rate, is used to establish the critical value in the
Bonferroni AEG test. The results from the cointegration tests based on the speci…cation in equation
(9) are given in Table 2.
10
Asymptotic critical values are used for both the standard Engle-Granger
test (denoted ¹1G) and the Bonferroni AEG test (denoted ¹1G
C
) and are provided in Table 2; the
¹1G
C
critical value is obtained from Table A3 and linear interpolation.
Table 2: Cointegration tests.
Test statistic -3.43
Critical value ¹1G
C
-3.47
Critical value ¹1G -3.34
Notes: Nominal size is 0.05.
As can be seen, the null hypothesis of no cointegration is rejected if the standard method is used,
as the test statistic is smaller than the critical value for the traditional AEG test. However, when
the Bonferroni AEG test is used, the null hypothesis is not rejected. Thus, performing inference using
robust methods, there is no strong evidence of cointegration, or co-movement, between the nominal
interest rate and in‡ation in U.S. data. This raises doubts about the validity of the Fisher hypothesis,
and also illustrates the importance of controlling for the unknown degree of persistence in the data;
assuming unit roots in the data, the cointegration test would have resulted in evidence favorable of
9
Lag length in the DF-GLS test was determined using the Schwarz (1978) information criterion. For the KPSS test,
a Newey-West estimator was employed to correct for serial correlation.
10
As in the DF-GLS test, lag length in the test equation is determined using the Schwarz (1978) criterion.
13
the Fisher hypothesis. Having looked at a traditional application from the macroeconomic literature,
we next turn to a recent issue from …nancial economics.
5.2 Consumption, aggregate wealth and stock returns
Many studies argue that …nancial valuation ratios such as the dividend- and earnings-price ratios
may have predictive power for excess stock returns over the risk-free rate. In a novel attempt to
tie macroeconomic variables more closely to …nancial markets, Lettau and Ludvigson (2001) argue
that consumption is a function of aggregate wealth. Based on this claim, they suggest that aggregate
consumption (/
t
), asset holdings (a
t
) and labour income (n
t
) are cointegrated and that the deviation
from equilibrium is useful in terms of predicting both excess stock returns and real stock returns.
The empirical speci…cation used by Lettau and Ludvigson accordingly takes its starting point in a
cointegrating relationship of the type
/
t
= j +0a
t
+`n
t
+.
t
. (10)
where the error term .
t
is assumed to be a stationary ARMA process which has predictive power for
future returns.
However, there is no strong a priori reason to assume that the above variables contain pure unit
roots.
11
We therefore investigate the sensitivity of Lettau and Ludvigson’s results when the uncertainty
regarding the persistence in the data is taken into account. Quarterly data on US consumption, asset
holdings and labour income ranging from the …rst quarter 1952 to the fourth quarter 2006 were obtained
from Professor Ludvigson’s web page;
12
all variables are given by the natural logarithm of real, per
capita data.
Table 3 shows the results from unit root tests and stationarity tests for all variables and also
provides the median unbiased estimates of c, ^ c, as well as 90 percent con…dence intervals.
13
The
11
As was shown above, the persistence of the dependent variable is of special importance when using the AEG test.
The assumption of a unit root in consumption is thus of particular interest. Although this conjecture …nds some support
– see, for example, Hall (1978) and Gali (1993) – the opinion in the litterature is far from unanimous. For instance,
the vast literatue that uses linear trends to detrend consumption – see, for example, Cooper and Ejarque (2000) and
Casares (2007) – implicitly or explicitly assumes that consumption is trend stationary rather than generated by a unit
root process. Furthermore, it has been argued that consumption and output should be integrated of the same order.
Thus, if output is trend stationary (e.g. Flavin, 1981 and Diebold and Senhadji, 1996) then consumption should be as
well.
12
http://www.econ.nyu.edu/user/ludvigsons/
13
Note that in this application, the unit root tests have both constant and trend included in the speci…cation. Thus,
the estimates and con…dence intervals of c are derived using the values in Table A2; again, linear interpolation is used.
14
evidence for unit roots in consumption and labour income seems strong, whereas it is mixed for asset
holdings.
Table 3: Unit root tests.
/
t
a
t
n
t
DF-GLS -1.95 -2.54 -0.78
KPSS 0.36

0.20

0.38

^ c -4.06 -9.98 2.32
90% CI for c [-12.28, 3.35] [-19.63, 2.32] [-2.18, 4.44]
Notes: * indicates signi…cance at the …ve percent level.
As in the previous application, we choose a signi…cance level of …ve percent for the cointegration
tests and set c
1
= 0.5. The results from the AEG and Bonferroni AEG cointegration tests are shown
in Table 4. The null hypothesis of no cointegration is rejected regardless of which test is used. The
robust cointegration methods developed here thus support the conclusion of Lettau and Ludvigson
that US consumption, asset holdings and labour income are cointegrated.
Table 4: Cointegration tests.
Test statistic -4.03
Critical value ¹1G
C
-3.86
Critical value ¹1G -3.77
Notes: Nominal size is 0.05.
6 Conclusion
For many economic time series, it is di¢cult to justify theoretically that they are generated by unit
root processes. This is problematic from an empirical point of view since cointegration tests may
be misleading when the data follow near-integrated, rather than pure unit root, processes. The size
distortions of cointegration tests relying on the unit root assumption – combined with the fact that
standard OLS inference could lead to spurious results – makes it unclear how to analyze a multivariate
time series of near-integrated variables.
15
In this paper, we have extended a standard residual-based cointegration test to allow for an unknown
local deviation from the unit root assumption. This more robust test is easy to implement and Monte
Carlo simulations show that it works well in …nite samples. Unlike standard cointegration tests, the
methods developed in this paper thus provide a means of performing valid inference on a multivariate
near unit root process. The framework suggested in this paper therefore takes another step towards
addressing the problems associated with inference when variables are near-integrated. The methods
presented here take their starting point in the work of Engle and Granger (1987). In future research
it would also be of interest to see Johansen’s (1988,1991) VAR-based framework extended to a setting
with near-integrated variables.
16
References
[1] Campbell, J.Y. and M. Yogo, 2006. E¢cient Tests of Stock Return Predictability, Journal of
Financial Economics 81, 27-60.
[2] Casares, M., 2007. The New Keynesian Model and the Euro Area Business Cycle, Oxford Bulletin
of Economics and Statistics 69, 209-244.
[3] Cavanagh, C., G. Elliot, and J. Stock, 1995. Inference in models with nearly integrated regressors,
Econometric Theory 11, 1131-1147.
[4] Cooper, R. and J. Ejarque, 2000. Financial Intermediation and Aggregate Fluctuations: A Quan-
titative Analysis, Macroeconomic Dynamics 4, 423-447.
[5] Crowder, W.J., and D.L. Ho¤man, 1996. The Long-Run Relationship between Nominal Interest
Rates and In‡ation: The Fisher Equation Revisited, Journal of Money, Credit and Banking 28,
102-118.
[6] Culver, S.E. and D.H. Papell, 1997. Is There a Unit Root in the In‡ation Rate? Evidence from
Sequential Break and Panel Data Models, Journal of Applied Econometrics 12, 435-444.
[7] Diebold, F.X. and A.S. Senhadji, 1996. The Uncertain Unit Root in Real GNP: Comment, Amer-
ican Economic Review 86, 1291-1298.
[8] Elliot, G., 1998. On the Robustness of Cointegration Methods When Regressors Almost Have
Unit Roots, Econometrica 66, 149-158.
[9] Elliot G., T.J. Rothenberg, and J.H. Stock, 1996. E¢cient Tests for an Autoregressive Unit Root,
Econometrica 64, 813-836.
[10] Engle, R.F. and C.W.J. Granger, 1987. Co-Integration and Error Correction: Representation,
Estimation, and Testing, Econometrica 55, 251-276.
[11] Evans, M. and K. Lewis, 1995. Do Expected Shifts in In‡ation A¤ect Estimates of the Long-Run
Fisher Relation?, Journal of Finance 50, 225-253.
[12] Flavin, M., 1981. The Adjustment of Consumption to Changing Expectations about Future In-
come, Journal of Political Economy 89, 974-1009.
17
[13] Gali, J., 1993. Variability of Durable and Nondurable Consumption: Evidence for Six O.E.C.D.
Countries, Review of Economics and Statistics 75, 418-428.
[14] Hall, R.E., 1978. Stochastic Implications of the Life-Cycle-Permanent-Income Hypothesis: Theory
and Evidence, Journal of Political Economy 86, 971-987.
[15] Hjalmarsson, E., and P. Österholm, 2007. Testing for Cointegration Using the Johansen Method-
ology when Variables are Near-Integrated, IMF Working Paper 07/141, International Monetary
Fund.
[16] Jansson, M., and M.J. Moreira, 2006. Optimal Inference in Regression Models with Nearly Inte-
grated Regressors, Econometrica 74, 681-714.
[17] Johansen, S., 1988. Statistical Analysis of Cointegration Vectors, Journal of Economic Dynamics
and Control 12, 231-254.
[18] Johansen, S., 1991. Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian
Vector Autoregressive Models, Econometrica 59, 1551-1580.
[19] Lettau, M. and S. Ludvigson, 2001. Consumption, Aggregate Wealth and Expected Stock Returns,
Journal of Finance 56, 815-849.
[20] Leybourne, S.J. and B.P.M. McCabe, 1993. A Simple Test for Cointegration, Oxford Bulletin of
Economics and Statistics 55, 97-103.
[21] Mishkin, F.S., 1992. Is the Fisher e¤ect for real?, Journal of Monetary Economics 30, 195-215.
[22] Phillips, P.C.B., 1987. Towards a Uni…ed Asymptotic Theory of Autoregression, Biometrika 74,
535-547.
[23] Phillips, P.C.B., 1988. Regression Theory for Near-Integrated Time Series, Econometrica 56,
1021-1043.
[24] Phillips, P.C.B., ans S. Ouliaris, 1990. Asymptotic Properties of Residual Based Tests for Coin-
tegration, Econometrica 58, 165-193.
[25] Phillips, P.C.B., and V. Solo, 1992. Asymptotics for Linear Processes, Annals of Statistics, 20,
971-1001.
18
[26] Schwarz, G., 1978. Estimating the Dimension of a Model, Annals of Statistics 6, 461–464.
[27] Shin, Y., 1994. A Residual-Based Test of the Null of Cointegration Against the Alternative of No
Cointegration, Econometric Theory 10, 91-115.
[28] Stock, J.H., 1991. Con…dence intervals for the largest autoregressive root in U.S. economic time-
series. Journal of Monetary Economics 28, 435-460.
[29] Stock, J.H., 1997. Cointegration, Long-Run Comovements, and Long-Horizon Forecasting, in
D. Kreps and K.F. Wallis (eds), Advances in Econometrics: Proceedings of the Seventh World
Congress of the Econometric Society, vol. III. Cambridge: Cambridge University Press, 34-60.
[30] Wallace, M. and J. Warner, 1993. The Fisher E¤ect and the Term Structure of Interest Rates:
Test of Cointegration. Review of Economics and Statistics 75, 320-324.
[31] Wright J.H., 2000. Con…dence Sets for Cointegrating Coe¢cients Based on Stationarity Tests,
Journal of Business and Economic Statistics 18, 211-222.
[32] Wu, J.-L. and S.-L. Chen, 2001. Mean Reversion of Interest Rates in the Eurocurrency Market,
Oxford Bulletin of Economics and Statistics 63, 459-474.
[33] Wu, Y. and H. Zhang, (1996). Mean Reversion in Interest Rates: New Evidence from a Panel of
OECD Countries, Journal of Money, Credit and Banking 28, 604-621.
19
T
a
b
l
e
A
1
:
L
o
w
e
r
c
o
n

d
e
n
c
e
b
o
u
n
d
s
f
o
r
c
b
a
s
e
d
o
n
t
h
e
D
F
-
G
L
S
s
t
a
t
i
s
t
i
c
.
F
o
r
a
g
i
v
e
n
v
a
l
u
e
o
f
t
h
e
D
F
-
G
L
S
s
t
a
t
i
s
t
i
c
,
w
i
t
h
o
u
t
a
t
i
m
e
t
r
e
n
d
i
n
c
l
u
d
e
d
,
t
h
e
f
o
l
l
o
w
i
n
g
c
o
l
u
m
n
s
g
i
v
e
l
o
w
e
r
c
o
n

d
e
n
c
e
b
o
u
n
d
s
o
f
t
h
e
l
o
c
a
l
-
t
o
-
u
n
i
t
y
p
a
r
a
m
e
t
e
r
c
w
i
t
h
c
o
n

d
e
n
c
e
l
e
v
e
l
s
o
f
9
5
.
7
5
.
5
0
.
2
5
.
a
n
d
5
p
e
r
c
e
n
t
,
r
e
s
p
e
c
t
i
v
e
l
y
.
D
F
-
G
L
S
9
5
%
7
5
%
5
0
%
2
5
%
5
%
D
F
-
G
L
S
9
5
%
7
5
%
5
0
%
2
5
%
5
%
1
.
0
-
0
.
2
9
0
.
7
2
1
.
4
7
2
.
3
9
4
.
2
3
-
2
.
0
-
1
4
.
9
0
-
1
0
.
4
1
-
7
.
4
4
-
4
.
5
1
-
0
.
3
1
0
.
9
-
0
.
4
0
0
.
6
5
1
.
4
1
2
.
3
4
4
.
1
9
-
2
.
1
-
1
5
.
9
5
-
1
1
.
3
7
-
8
.
2
5
-
5
.
2
1
-
0
.
8
8
0
.
8
-
0
.
5
0
0
.
5
7
1
.
3
5
2
.
2
9
4
.
1
5
-
2
.
2
-
1
7
.
1
4
-
1
2
.
3
5
-
9
.
0
9
-
5
.
9
4
-
1
.
3
6
0
.
7
-
0
.
6
3
0
.
4
9
1
.
2
9
2
.
2
4
4
.
1
2
-
2
.
3
-
1
8
.
3
4
-
1
3
.
3
8
-
9
.
9
7
-
6
.
7
4
-
1
.
9
8
0
.
6
-
0
.
7
6
0
.
4
0
1
.
2
3
2
.
1
9
4
.
0
8
-
2
.
4
-
1
9
.
5
7
-
1
4
.
4
2
-
1
0
.
9
2
-
7
.
4
8
-
2
.
5
5
0
.
5
-
0
.
9
1
0
.
3
0
1
.
1
5
2
.
1
3
4
.
0
4
-
2
.
5
-
2
0
.
8
4
-
1
5
.
4
8
-
1
1
.
8
9
-
8
.
3
2
-
3
.
2
8
0
.
4
-
1
.
0
7
0
.
2
0
1
.
0
7
2
.
0
7
4
.
0
0
-
2
.
6
-
2
2
.
1
5
-
1
6
.
6
1
-
1
2
.
9
1
-
9
.
1
9
-
3
.
9
5
0
.
3
-
1
.
2
5
0
.
0
9
0
.
9
9
2
.
0
2
3
.
9
5
-
2
.
7
-
2
3
.
5
3
-
1
7
.
7
8
-
1
3
.
9
5
-
1
0
.
0
6
-
4
.
6
9
0
.
2
-
1
.
4
6
-
0
.
0
3
0
.
9
0
1
.
9
4
3
.
9
0
-
2
.
8
-
2
4
.
9
3
-
1
8
.
9
8
-
1
5
.
0
3
-
1
1
.
0
6
-
5
.
5
2
0
.
1
-
1
.
6
6
-
0
.
1
7
0
.
8
0
1
.
8
7
3
.
8
5
-
2
.
9
-
2
6
.
3
4
-
2
0
.
2
0
-
1
6
.
1
3
-
1
2
.
0
6
-
6
.
2
8
0
.
0
-
1
.
8
9
-
0
.
3
1
0
.
7
0
1
.
7
9
3
.
8
0
-
3
.
0
-
2
7
.
7
1
-
2
1
.
4
9
-
1
7
.
2
9
-
1
3
.
0
8
-
7
.
1
4
-
0
.
1
-
2
.
1
4
-
0
.
4
6
0
.
5
9
1
.
7
1
3
.
7
5
-
3
.
1
-
2
9
.
2
7
-
2
2
.
8
1
-
1
8
.
4
5
-
1
4
.
1
2
-
7
.
9
7
-
0
.
2
-
2
.
4
1
-
0
.
6
3
0
.
4
8
1
.
6
2
3
.
6
9
-
3
.
2
-
3
0
.
8
6
-
2
4
.
1
7
-
1
9
.
6
2
-
1
5
.
2
1
-
8
.
8
8
-
0
.
3
-
2
.
7
2
-
0
.
8
2
0
.
3
4
1
.
5
3
3
.
6
3
-
3
.
3
-
3
2
.
4
4
-
2
5
.
5
3
-
2
0
.
8
7
-
1
6
.
3
3
-
9
.
8
4
-
0
.
4
-
3
.
0
5
-
1
.
0
3
0
.
1
8
1
.
4
2
3
.
5
7
-
3
.
4
-
3
4
.
0
6
-
2
6
.
9
4
-
2
2
.
1
5
-
1
7
.
5
2
-
1
0
.
8
3
-
0
.
5
-
3
.
4
5
-
1
.
2
9
-
0
.
0
2
1
.
3
0
3
.
5
1
-
3
.
5
-
3
5
.
7
8
-
2
8
.
3
9
-
2
3
.
4
9
-
1
8
.
7
0
-
1
1
.
8
0
-
0
.
6
-
3
.
8
4
-
1
.
6
0
-
0
.
2
3
1
.
1
5
3
.
4
0
-
3
.
6
-
3
7
.
4
3
-
2
9
.
8
7
-
2
4
.
8
5
-
1
9
.
8
7
-
1
2
.
8
7
-
0
.
7
-
4
.
3
1
-
1
.
9
4
-
0
.
4
7
0
.
9
8
3
.
2
8
-
3
.
7
-
3
9
.
0
9
-
3
1
.
4
4
-
2
6
.
2
4
-
2
1
.
1
6
-
1
3
.
9
6
-
0
.
8
-
4
.
8
7
-
2
.
3
2
-
0
.
7
5
0
.
7
8
3
.
1
7
-
3
.
8
-
4
0
.
8
5
-
3
2
.
9
8
-
2
7
.
6
5
-
2
2
.
4
8
-
1
5
.
0
9
-
0
.
9
-
5
.
4
4
-
2
.
7
8
-
1
.
1
0
0
.
5
4
3
.
0
6
-
3
.
9
-
4
2
.
6
9
-
3
4
.
5
5
-
2
9
.
1
1
-
2
3
.
8
2
-
1
6
.
2
5
-
1
.
0
-
6
.
0
4
-
3
.
2
7
-
1
.
4
7
0
.
2
8
2
.
9
1
-
4
.
0
-
4
4
.
5
2
-
3
6
.
2
2
-
3
0
.
6
2
-
2
5
.
1
8
-
1
7
.
5
2
-
1
.
1
-
6
.
7
3
-
3
.
7
9
-
1
.
9
0
-
0
.
0
6
2
.
7
1
-
4
.
1
-
4
6
.
3
5
-
3
7
.
8
7
-
3
2
.
1
7
-
2
6
.
5
5
-
1
8
.
7
1
-
1
.
2
-
7
.
4
5
-
4
.
3
7
-
2
.
3
5
-
0
.
3
9
2
.
4
9
-
4
.
2
-
4
8
.
2
4
-
3
9
.
5
0
-
3
3
.
7
0
-
2
7
.
9
3
-
1
9
.
8
7
-
1
.
3
-
8
.
1
9
-
4
.
9
7
-
2
.
8
8
-
0
.
7
6
2
.
2
9
-
4
.
3
-
5
0
.
1
4
-
4
1
.
2
7
-
3
5
.
3
1
-
2
9
.
4
4
-
2
1
.
2
2
-
1
.
4
-
9
.
0
4
-
5
.
6
6
-
3
.
4
0
-
1
.
1
8
2
.
0
1
-
4
.
4
-
5
2
.
1
4
-
4
3
.
0
7
-
3
6
.
9
4
-
3
0
.
9
4
-
2
2
.
5
7
-
1
.
5
-
9
.
9
0
-
6
.
3
3
-
3
.
9
7
-
1
.
6
5
1
.
7
4
-
4
.
5
-
5
3
.
9
6
-
4
4
.
8
6
-
3
8
.
5
8
-
3
2
.
4
5
-
2
3
.
8
9
-
1
.
6
-
1
0
.
8
2
-
7
.
0
5
-
4
.
6
0
-
2
.
1
5
1
.
3
8
-
4
.
6
-
5
6
.
0
8
-
4
6
.
6
8
-
4
0
.
2
3
-
3
4
.
0
0
-
2
5
.
2
1
-
1
.
7
-
1
1
.
7
5
-
7
.
8
5
-
5
.
2
3
-
2
.
7
2
1
.
0
3
-
4
.
7
-
5
8
.
2
0
-
4
8
.
5
4
-
4
1
.
9
5
-
3
5
.
6
7
-
2
6
.
5
9
-
1
.
8
-
1
2
.
7
8
-
8
.
6
5
-
5
.
9
4
-
3
.
2
7
0
.
5
9
-
4
.
8
-
6
0
.
2
7
-
5
0
.
3
9
-
4
3
.
7
0
-
3
7
.
2
9
-
2
8
.
0
5
-
1
.
9
-
1
3
.
8
4
-
9
.
5
1
-
6
.
6
9
-
3
.
8
6
0
.
2
2
-
4
.
9
-
6
2
.
3
8
-
5
2
.
3
1
-
4
5
.
5
0
-
3
8
.
9
0
-
2
9
.
5
3
20
T
a
b
l
e
A
2
:
L
o
w
e
r
c
o
n

d
e
n
c
e
b
o
u
n
d
s
f
o
r
c
b
a
s
e
d
o
n
t
h
e
D
F
-
G
L
S
s
t
a
t
i
s
t
i
c
w
i
t
h
a
l
i
n
e
a
r
t
i
m
e
t
r
e
n
d
i
n
c
l
u
d
e
d
.
F
o
r
a
g
i
v
e
n
v
a
l
u
e
o
f
t
h
e
D
F
-
G
L
S
s
t
a
t
i
s
t
i
c
,
a
l
l
o
w
i
n
g
f
o
r
a
l
i
n
e
a
r
t
i
m
e
t
r
e
n
d
,
t
h
e
f
o
l
l
o
w
i
n
g
c
o
l
u
m
n
s
g
i
v
e
l
o
w
e
r
c
o
n

d
e
n
c
e
b
o
u
n
d
s
o
f
t
h
e
l
o
c
a
l
-
t
o
-
u
n
i
t
y
p
a
r
a
m
e
t
e
r
c
w
i
t
h
c
o
n

d
e
n
c
e
l
e
v
e
l
s
o
f
9
5
.
7
5
.
5
0
.
2
5
.
a
n
d
5
p
e
r
c
e
n
t
,
r
e
s
p
e
c
t
i
v
e
l
y
.
D
F
-
G
L
S
9
5
%
7
5
%
5
0
%
2
5
%
5
%
D
F
-
G
L
S
9
5
%
7
5
%
5
0
%
2
5
%
5
%
1
.
0
2
.
2
0
2
.
6
3
3
.
0
7
3
.
7
2
5
.
2
4
-
2
.
0
-
1
2
.
9
0
-
8
.
1
5
-
4
.
5
6
1
.
6
9
3
.
2
7
0
.
9
2
.
1
6
2
.
6
0
3
.
0
4
3
.
6
9
5
.
2
0
-
2
.
1
-
1
4
.
0
4
-
9
.
1
5
-
5
.
5
4
1
.
2
8
3
.
1
2
0
.
8
2
.
1
2
2
.
5
7
3
.
0
1
3
.
6
5
5
.
1
6
-
2
.
2
-
1
5
.
2
6
-
1
0
.
1
7
-
6
.
4
8
-
2
.
1
0
2
.
9
5
0
.
7
2
.
0
9
2
.
5
3
2
.
9
7
3
.
6
2
5
.
1
3
-
2
.
3
-
1
6
.
5
2
-
1
1
.
2
8
-
7
.
4
7
-
3
.
2
2
2
.
7
6
0
.
6
2
.
0
5
2
.
5
0
2
.
9
3
3
.
5
8
5
.
0
9
-
2
.
4
-
1
7
.
8
5
-
1
2
.
4
0
-
8
.
4
9
-
4
.
2
6
2
.
5
8
0
.
5
2
.
0
2
2
.
4
6
2
.
9
0
3
.
5
5
5
.
0
5
-
2
.
5
-
1
9
.
1
4
-
1
3
.
5
5
-
9
.
5
9
-
5
.
3
0
2
.
3
9
0
.
4
1
.
9
7
2
.
4
2
2
.
8
6
3
.
5
1
5
.
0
1
-
2
.
6
-
2
0
.
4
9
-
1
4
.
7
7
-
1
0
.
6
7
-
6
.
3
3
2
.
1
9
0
.
3
1
.
9
3
2
.
3
8
2
.
8
2
3
.
4
7
4
.
9
7
-
2
.
7
-
2
1
.
9
7
-
1
6
.
0
4
-
1
1
.
8
0
-
7
.
4
1
1
.
9
6
0
.
2
1
.
8
8
2
.
3
4
2
.
7
8
3
.
4
2
4
.
9
3
-
2
.
8
-
2
3
.
4
4
-
1
7
.
3
5
-
1
2
.
9
8
-
8
.
4
7
1
.
6
1
0
.
1
1
.
8
3
2
.
3
0
2
.
7
4
3
.
3
8
4
.
8
8
-
2
.
9
-
2
4
.
9
7
-
1
8
.
6
7
-
1
4
.
2
0
-
9
.
6
2
-
1
.
5
5
0
.
0
1
.
7
8
2
.
2
6
2
.
7
0
3
.
3
3
4
.
8
4
-
3
.
0
-
2
6
.
5
5
-
2
0
.
0
2
-
1
5
.
4
7
-
1
0
.
7
5
-
3
.
1
0
-
0
.
1
1
.
7
2
2
.
2
2
2
.
6
5
3
.
2
9
4
.
7
9
-
3
.
1
-
2
8
.
1
4
-
2
1
.
4
8
-
1
6
.
7
8
-
1
1
.
9
1
-
4
.
2
7
-
0
.
2
1
.
6
4
2
.
1
7
2
.
6
1
3
.
2
4
4
.
7
5
-
3
.
2
-
2
9
.
8
6
-
2
2
.
9
7
-
1
8
.
1
0
-
1
3
.
1
9
-
5
.
5
5
-
0
.
3
1
.
5
6
2
.
1
2
2
.
5
6
3
.
2
0
4
.
7
0
-
3
.
3
-
3
1
.
6
4
-
2
4
.
4
9
-
1
9
.
5
1
-
1
4
.
4
8
-
6
.
6
8
-
0
.
4
1
.
4
7
2
.
0
7
2
.
5
2
3
.
1
5
4
.
6
4
-
3
.
4
-
3
3
.
4
2
-
2
6
.
0
5
-
2
0
.
9
6
-
1
5
.
8
0
-
7
.
9
1
-
0
.
5
1
.
3
2
2
.
0
2
2
.
4
7
3
.
1
0
4
.
5
9
-
3
.
5
-
3
5
.
2
1
-
2
7
.
6
7
-
2
2
.
4
5
-
1
7
.
1
5
-
9
.
1
2
-
0
.
6
-
0
.
8
1
1
.
9
5
2
.
4
2
3
.
0
5
4
.
5
4
-
3
.
6
-
3
7
.
0
9
-
2
9
.
3
7
-
2
3
.
9
5
-
1
8
.
5
3
-
1
0
.
3
0
-
0
.
7
-
1
.
5
8
1
.
8
9
2
.
3
6
3
.
0
1
4
.
4
9
-
3
.
7
-
3
8
.
9
9
-
3
1
.
0
9
-
2
5
.
5
6
-
1
9
.
9
3
-
1
1
.
6
2
-
0
.
8
-
2
.
2
9
1
.
8
2
2
.
3
1
2
.
9
5
4
.
4
3
-
3
.
8
-
4
0
.
9
7
-
3
2
.
8
5
-
2
7
.
1
9
-
2
1
.
4
2
-
1
2
.
9
6
-
0
.
9
-
2
.
9
5
1
.
7
5
2
.
2
6
2
.
8
9
4
.
3
7
-
3
.
9
-
4
3
.
0
6
-
3
4
.
6
4
-
2
8
.
8
5
-
2
2
.
9
7
-
1
4
.
3
4
-
1
.
0
-
3
.
7
0
1
.
6
1
2
.
1
8
2
.
8
2
4
.
3
1
-
4
.
0
-
4
5
.
1
8
-
3
6
.
5
0
-
3
0
.
5
6
-
2
4
.
5
7
-
1
5
.
7
9
-
1
.
1
-
4
.
4
3
1
.
4
5
2
.
1
0
2
.
7
6
4
.
2
5
-
4
.
1
-
4
7
.
1
8
-
3
8
.
4
5
-
3
2
.
3
4
-
2
6
.
1
8
-
1
7
.
3
1
-
1
.
2
-
5
.
1
5
-
0
.
7
2
2
.
0
3
2
.
6
9
4
.
1
7
-
4
.
2
-
4
9
.
3
6
-
4
0
.
3
5
-
3
4
.
1
3
-
2
7
.
8
9
-
1
8
.
7
7
-
1
.
3
-
6
.
0
1
-
1
.
8
5
1
.
9
2
2
.
6
0
4
.
0
9
-
4
.
3
-
5
1
.
6
6
-
4
2
.
3
7
-
3
6
.
0
1
-
2
9
.
5
6
-
2
0
.
1
9
-
1
.
4
-
6
.
8
3
-
2
.
7
5
1
.
8
0
2
.
5
2
4
.
0
1
-
4
.
4
-
5
3
.
9
1
-
4
4
.
4
6
-
3
7
.
9
0
-
3
1
.
3
1
-
2
1
.
8
3
-
1
.
5
-
7
.
7
4
-
3
.
6
2
1
.
6
3
2
.
4
2
3
.
9
1
-
4
.
5
-
5
6
.
2
7
-
4
6
.
6
0
-
3
9
.
8
3
-
3
3
.
1
5
-
2
3
.
4
4
-
1
.
6
-
8
.
6
9
-
4
.
4
6
1
.
3
6
2
.
3
1
3
.
8
1
-
4
.
6
-
5
8
.
7
4
-
4
8
.
7
4
-
4
1
.
8
9
-
3
5
.
0
4
-
2
5
.
0
0
-
1
.
7
-
9
.
6
7
-
5
.
3
3
-
1
.
5
6
2
.
1
9
3
.
6
9
-
4
.
7
-
6
1
.
2
0
-
5
0
.
9
8
-
4
3
.
9
4
-
3
6
.
9
9
-
2
6
.
6
6
-
1
.
8
-
1
0
.
6
5
-
6
.
2
2
-
2
.
6
9
2
.
0
6
3
.
5
6
-
4
.
8
-
6
3
.
7
8
-
5
3
.
3
2
-
4
6
.
0
7
-
3
8
.
9
6
-
2
8
.
5
2
-
1
.
9
-
1
1
.
7
6
-
7
.
1
7
-
3
.
6
4
1
.
8
9
3
.
4
2
-
4
.
9
-
6
6
.
2
5
-
5
5
.
6
4
-
4
8
.
2
9
-
4
0
.
9
5
-
3
0
.
2
8
21
T
a
b
l
e
A
3
:
F
i
v
e
p
e
r
c
e
n
t
c
r
i
t
i
c
a
l
v
a
l
u
e
s
f
o
r
t
h
e
¹
1
G
s
t
a
t
i
s
t
i
c
.
T
h
i
s
t
a
b
l
e
g
i
v
e
s
t
h
e
c
r
i
t
i
c
a
l
v
a
l
u
e
s
f
o
r
t
h
e
¹
1
G
s
t
a
t
i
s
t
i
c
a
t
t
h
e

v
e
p
e
r
c
e
n
t
l
e
v
e
l
,
f
o
r
d
i
¤
e
r
e
n
t
v
a
l
u
e
s
o
f
c
u
n
d
e
r
t
h
e
a
s
s
u
m
p
t
i
o
n
t
h
a
t
c
1
=
.
.
.
=
c
m
=
c
,
a
n
d
f
o
r
o
n
e
t
o

v
e
r
e
g
r
e
s
s
o
r
s
.
T
h
e

r
s
t
s
e
t
o
f
v
a
l
u
e
s
p
r
o
v
i
d
e
t
h
e
c
r
i
t
i
c
a
l
v
a
l
u
e
s
w
h
e
n
n
o
i
n
t
e
r
c
e
p
t
i
s
i
n
c
l
u
d
e
d
i
n
t
h
e
c
o
i
n
t
e
g
r
a
t
i
n
g
r
e
g
r
e
s
s
i
o
n
.
T
h
e
s
e
c
o
n
d
s
e
t
p
r
o
v
i
d
e
s
t
h
e
v
a
l
u
e
s
w
h
e
n
a
n
i
n
t
e
r
c
e
p
t
,
b
u
t
n
o
t
i
m
e
t
r
e
n
d
i
s
i
n
c
l
u
d
e
d
a
n
d
t
h
e
t
h
i
r
d
s
e
t
o
f
v
a
l
u
e
s
r
e
p
r
e
s
e
n
t
t
h
e
c
a
s
e
w
i
t
h
b
o
t
h
a
n
i
n
t
e
r
c
e
p
t
a
n
d
a
l
i
n
e
a
r
t
i
m
e
t
r
e
n
d
.
T
h
e
v
a
l
u
e
s
a
r
e
b
a
s
e
d
o
n
1
0
0
.
0
0
0
r
e
p
e
t
i
t
i
o
n
s
w
i
t
h
T
=
1
.
0
0
0
.
N
o
c
o
n
s
t
a
n
t
C
o
n
s
t
a
n
t
C
o
n
s
t
a
n
t
a
n
d
t
r
e
n
d
c
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
0
-
2
.
7
7
-
3
.
3
0
-
3
.
7
3
-
4
.
0
9
-
4
.
4
1
-
3
.
3
4
-
3
.
7
7
-
4
.
1
0
-
4
.
4
2
-
4
.
7
2
-
3
.
7
9
-
4
.
1
4
-
4
.
4
4
-
4
.
7
3
-
5
.
0
0
-
1
-
2
.
8
0
-
3
.
3
2
-
3
.
7
4
-
4
.
0
9
-
4
.
4
2
-
3
.
3
7
-
3
.
7
6
-
4
.
1
2
-
4
.
4
3
-
4
.
7
3
-
3
.
7
9
-
4
.
1
4
-
4
.
4
6
-
4
.
7
2
-
5
.
0
1
-
2
-
2
.
8
8
-
3
.
3
5
-
3
.
7
5
-
4
.
1
0
-
4
.
4
2
-
3
.
4
0
-
3
.
7
8
-
4
.
1
2
-
4
.
4
4
-
4
.
7
3
-
3
.
8
2
-
4
.
1
6
-
4
.
4
5
-
4
.
7
4
-
5
.
0
1
-
3
-
2
.
9
6
-
3
.
3
9
-
3
.
7
8
-
4
.
1
2
-
4
.
4
4
-
3
.
4
5
-
3
.
8
2
-
4
.
1
5
-
4
.
4
6
-
4
.
7
5
-
3
.
8
6
-
4
.
1
8
-
4
.
4
8
-
4
.
7
6
-
5
.
0
3
-
4
-
3
.
0
5
-
3
.
4
5
-
3
.
8
2
-
4
.
1
5
-
4
.
4
5
-
3
.
5
0
-
3
.
8
6
-
4
.
1
7
-
4
.
4
7
-
4
.
7
6
-
3
.
8
9
-
4
.
2
1
-
4
.
5
0
-
4
.
7
8
-
5
.
0
3
-
5
-
3
.
1
5
-
3
.
5
1
-
3
.
8
6
-
4
.
1
7
-
4
.
4
6
-
3
.
5
6
-
3
.
8
9
-
4
.
2
1
-
4
.
4
9
-
4
.
7
7
-
3
.
9
4
-
4
.
2
4
-
4
.
5
3
-
4
.
7
9
-
5
.
0
5
-
6
-
3
.
2
3
-
3
.
5
8
-
3
.
9
1
-
4
.
2
1
-
4
.
5
1
-
3
.
6
2
-
3
.
9
4
-
4
.
2
4
-
4
.
5
3
-
4
.
8
0
-
3
.
9
8
-
4
.
2
7
-
4
.
5
6
-
4
.
8
1
-
5
.
0
7
-
7
-
3
.
3
2
-
3
.
6
4
-
3
.
9
6
-
4
.
2
6
-
4
.
5
3
-
3
.
6
8
-
4
.
0
0
-
4
.
2
8
-
4
.
5
6
-
4
.
8
1
-
4
.
0
3
-
4
.
3
2
-
4
.
5
8
-
4
.
8
5
-
5
.
0
9
-
8
-
3
.
4
1
-
3
.
7
2
-
4
.
0
1
-
4
.
3
0
-
4
.
5
7
-
3
.
7
5
-
4
.
0
5
-
4
.
3
3
-
4
.
6
0
-
4
.
8
6
-
4
.
0
8
-
4
.
3
6
-
4
.
6
3
-
4
.
8
9
-
5
.
1
1
-
9
-
3
.
5
0
-
3
.
7
9
-
4
.
0
7
-
4
.
3
4
-
4
.
6
1
-
3
.
8
2
-
4
.
1
1
-
4
.
3
7
-
4
.
6
4
-
4
.
8
9
-
4
.
1
4
-
4
.
4
1
-
4
.
6
6
-
4
.
9
1
-
5
.
1
6
-
1
0
-
3
.
5
8
-
3
.
8
6
-
4
.
1
3
-
4
.
4
0
-
4
.
6
5
-
3
.
8
9
-
4
.
1
6
-
4
.
4
3
-
4
.
6
8
-
4
.
9
2
-
4
.
1
9
-
4
.
4
4
-
4
.
7
1
-
4
.
9
3
-
5
.
1
8
-
1
1
-
3
.
6
8
-
3
.
9
3
-
4
.
1
9
-
4
.
4
5
-
4
.
6
9
-
3
.
9
7
-
4
.
2
2
-
4
.
4
7
-
4
.
7
2
-
4
.
9
5
-
4
.
2
6
-
4
.
5
1
-
4
.
7
4
-
4
.
9
8
-
5
.
2
1
-
1
2
-
3
.
7
5
-
4
.
0
1
-
4
.
2
6
-
4
.
5
0
-
4
.
7
3
-
4
.
0
3
-
4
.
2
9
-
4
.
5
2
-
4
.
7
6
-
4
.
9
9
-
4
.
3
2
-
4
.
5
6
-
4
.
7
9
-
5
.
0
2
-
5
.
2
3
-
1
3
-
3
.
8
4
-
4
.
0
8
-
4
.
3
3
-
4
.
5
6
-
4
.
7
8
-
4
.
1
0
-
4
.
3
4
-
4
.
5
8
-
4
.
8
1
-
5
.
0
3
-
4
.
3
7
-
4
.
6
0
-
4
.
8
4
-
5
.
0
6
-
5
.
2
8
-
1
4
-
3
.
9
2
-
4
.
1
6
-
4
.
3
8
-
4
.
6
0
-
4
.
8
3
-
4
.
1
8
-
4
.
4
1
-
4
.
6
4
-
4
.
8
5
-
5
.
0
7
-
4
.
4
4
-
4
.
6
6
-
4
.
8
8
-
5
.
0
9
-
5
.
3
0
-
1
5
-
4
.
0
1
-
4
.
2
3
-
4
.
4
4
-
4
.
6
6
-
4
.
8
8
-
4
.
2
5
-
4
.
4
7
-
4
.
6
9
-
4
.
9
0
-
5
.
1
1
-
4
.
5
0
-
4
.
7
1
-
4
.
9
3
-
5
.
1
4
-
5
.
3
4
-
1
6
-
4
.
0
6
-
4
.
3
0
-
4
.
5
0
-
4
.
7
2
-
4
.
9
4
-
4
.
3
0
-
4
.
5
4
-
4
.
7
4
-
4
.
9
6
-
5
.
1
6
-
4
.
5
5
-
4
.
7
8
-
4
.
9
7
-
5
.
1
9
-
5
.
3
9
-
1
7
-
4
.
1
5
-
4
.
3
7
-
4
.
5
7
-
4
.
7
8
-
4
.
9
8
-
4
.
3
8
-
4
.
6
0
-
4
.
8
0
-
5
.
0
0
-
5
.
2
0
-
4
.
6
1
-
4
.
8
2
-
5
.
0
2
-
5
.
2
3
-
5
.
4
2
-
1
8
-
4
.
2
2
-
4
.
4
3
-
4
.
6
3
-
4
.
8
3
-
5
.
0
3
-
4
.
4
4
-
4
.
6
6
-
4
.
8
5
-
5
.
0
5
-
5
.
2
5
-
4
.
6
7
-
4
.
8
8
-
5
.
0
7
-
5
.
2
7
-
5
.
4
6
-
1
9
-
4
.
2
9
-
4
.
5
0
-
4
.
6
9
-
4
.
8
9
-
5
.
0
7
-
4
.
5
0
-
4
.
7
2
-
4
.
9
0
-
5
.
1
1
-
5
.
2
9
-
4
.
7
3
-
4
.
9
4
-
5
.
1
2
-
5
.
3
2
-
5
.
5
0
-
2
0
-
4
.
3
7
-
4
.
5
7
-
4
.
7
6
-
4
.
9
4
-
5
.
1
3
-
4
.
5
8
-
4
.
7
7
-
4
.
9
7
-
5
.
1
6
-
5
.
3
4
-
4
.
7
9
-
4
.
9
9
-
5
.
1
8
-
5
.
3
7
-
5
.
5
5
-
2
1
-
4
.
4
4
-
4
.
6
2
-
4
.
8
1
-
5
.
0
0
-
5
.
1
9
-
4
.
6
5
-
4
.
8
3
-
5
.
0
1
-
5
.
2
1
-
5
.
3
9
-
4
.
8
5
-
5
.
0
4
-
5
.
2
1
-
5
.
4
1
-
5
.
5
9
-
2
2
-
4
.
5
1
-
4
.
6
9
-
4
.
8
7
-
5
.
0
4
-
5
.
2
3
-
4
.
7
0
-
4
.
8
9
-
5
.
0
7
-
5
.
2
5
-
5
.
4
2
-
4
.
9
1
-
5
.
0
8
-
5
.
2
7
-
5
.
4
5
-
5
.
6
2
-
2
3
-
4
.
5
7
-
4
.
7
6
-
4
.
9
3
-
5
.
1
1
-
5
.
2
8
-
4
.
7
7
-
4
.
9
6
-
5
.
1
3
-
5
.
3
0
-
5
.
4
7
-
4
.
9
7
-
5
.
1
5
-
5
.
3
3
-
5
.
4
9
-
5
.
6
7
-
2
4
-
4
.
6
5
-
4
.
8
1
-
5
.
0
0
-
5
.
1
6
-
5
.
3
4
-
4
.
8
3
-
5
.
0
0
-
5
.
1
9
-
5
.
3
5
-
5
.
5
2
-
5
.
0
3
-
5
.
1
9
-
5
.
3
8
-
5
.
5
4
-
5
.
7
2
-
2
5
-
4
.
7
1
-
4
.
8
8
-
5
.
0
4
-
5
.
2
2
-
5
.
3
9
-
4
.
8
9
-
5
.
0
6
-
5
.
2
3
-
5
.
4
1
-
5
.
5
7
-
5
.
0
8
-
5
.
2
5
-
5
.
4
2
-
5
.
5
9
-
5
.
7
6
-
2
6
-
4
.
7
7
-
4
.
9
3
-
5
.
1
1
-
5
.
2
7
-
5
.
4
4
-
4
.
9
5
-
5
.
1
1
-
5
.
2
9
-
5
.
4
5
-
5
.
6
2
-
5
.
1
4
-
5
.
2
9
-
5
.
4
7
-
5
.
6
4
-
5
.
8
1
-
2
7
-
4
.
8
4
-
5
.
0
0
-
5
.
1
6
-
5
.
3
2
-
5
.
4
9
-
5
.
0
1
-
5
.
1
8
-
5
.
3
4
-
5
.
5
0
-
5
.
6
7
-
5
.
1
9
-
5
.
3
6
-
5
.
5
3
-
5
.
6
9
-
5
.
8
5
-
2
8
-
4
.
9
0
-
5
.
0
6
-
5
.
2
1
-
5
.
3
8
-
5
.
5
3
-
5
.
0
8
-
5
.
2
3
-
5
.
3
9
-
5
.
5
5
-
5
.
7
0
-
5
.
2
5
-
5
.
4
1
-
5
.
5
6
-
5
.
7
2
-
5
.
8
8
-
2
9
-
4
.
9
6
-
5
.
1
2
-
5
.
2
8
-
5
.
4
3
-
5
.
5
9
-
5
.
1
3
-
5
.
2
9
-
5
.
4
5
-
5
.
6
0
-
5
.
7
6
-
5
.
3
0
-
5
.
4
7
-
5
.
6
3
-
5
.
7
7
-
5
.
9
3
-
3
0
-
5
.
0
2
-
5
.
1
8
-
5
.
3
2
-
5
.
4
7
-
5
.
6
4
-
5
.
1
9
-
5
.
3
4
-
5
.
4
9
-
5
.
6
4
-
5
.
8
1
-
5
.
3
6
-
5
.
5
2
-
5
.
6
6
-
5
.
8
2
-
5
.
9
8
22
T
a
b
l
e
A
3
:
C
r
i
t
i
c
a
l
v
a
l
u
e
s
f
o
r
t
h
e
¹
1
G
s
t
a
t
i
s
t
i
c
(
c
o
n
t
i
n
u
e
d
)
.
N
o
c
o
n
s
t
a
n
t
C
o
n
s
t
a
n
t
C
o
n
s
t
a
n
t
a
n
d
t
r
e
n
d
c
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
-
3
1
-
5
.
0
9
-
5
.
2
3
-
5
.
3
9
-
5
.
5
3
-
5
.
6
9
-
5
.
2
5
-
5
.
3
9
-
5
.
5
5
-
5
.
7
0
-
5
.
8
5
-
5
.
4
1
-
5
.
5
7
-
5
.
7
2
-
5
.
8
6
-
6
.
0
2
-
3
2
-
5
.
1
4
-
5
.
3
0
-
5
.
4
4
-
5
.
5
9
-
5
.
7
4
-
5
.
3
0
-
5
.
4
6
-
5
.
6
1
-
5
.
7
5
-
5
.
9
0
-
5
.
4
7
-
5
.
6
2
-
5
.
7
6
-
5
.
9
2
-
6
.
0
7
-
3
3
-
5
.
2
0
-
5
.
3
5
-
5
.
4
9
-
5
.
6
3
-
5
.
7
9
-
5
.
3
6
-
5
.
5
1
-
5
.
6
6
-
5
.
8
0
-
5
.
9
4
-
5
.
5
2
-
5
.
6
7
-
5
.
8
2
-
5
.
9
5
-
6
.
1
1
-
3
4
-
5
.
2
6
-
5
.
4
1
-
5
.
5
4
-
5
.
6
9
-
5
.
8
3
-
5
.
4
2
-
5
.
5
6
-
5
.
7
0
-
5
.
8
5
-
5
.
9
9
-
5
.
5
7
-
5
.
7
2
-
5
.
8
6
-
6
.
0
1
-
6
.
1
5
-
3
5
-
5
.
3
2
-
5
.
4
6
-
5
.
6
0
-
5
.
7
5
-
5
.
8
9
-
5
.
4
7
-
5
.
6
1
-
5
.
7
6
-
5
.
9
0
-
6
.
0
4
-
5
.
6
2
-
5
.
7
6
-
5
.
9
1
-
6
.
0
5
-
6
.
2
0
-
3
6
-
5
.
3
8
-
5
.
5
1
-
5
.
6
5
-
5
.
8
0
-
5
.
9
4
-
5
.
5
3
-
5
.
6
6
-
5
.
8
1
-
5
.
9
5
-
6
.
0
9
-
5
.
6
7
-
5
.
8
2
-
5
.
9
6
-
6
.
1
0
-
6
.
2
4
-
3
7
-
5
.
4
4
-
5
.
5
6
-
5
.
7
1
-
5
.
8
4
-
5
.
9
8
-
5
.
5
8
-
5
.
7
1
-
5
.
8
6
-
5
.
9
9
-
6
.
1
3
-
5
.
7
2
-
5
.
8
7
-
6
.
0
1
-
6
.
1
4
-
6
.
2
8
-
3
8
-
5
.
5
0
-
5
.
6
3
-
5
.
7
6
-
5
.
9
0
-
6
.
0
3
-
5
.
6
4
-
5
.
7
7
-
5
.
9
1
-
6
.
0
4
-
6
.
1
8
-
5
.
7
9
-
5
.
9
2
-
6
.
0
6
-
6
.
1
9
-
6
.
3
3
-
3
9
-
5
.
5
5
-
5
.
6
8
-
5
.
8
1
-
5
.
9
4
-
6
.
0
7
-
5
.
6
9
-
5
.
8
2
-
5
.
9
5
-
6
.
0
9
-
6
.
2
2
-
5
.
8
4
-
5
.
9
7
-
6
.
1
0
-
6
.
2
4
-
6
.
3
6
-
4
0
-
5
.
6
0
-
5
.
7
3
-
5
.
8
6
-
5
.
9
9
-
6
.
1
2
-
5
.
7
4
-
5
.
8
7
-
6
.
0
0
-
6
.
1
3
-
6
.
2
7
-
5
.
8
8
-
6
.
0
1
-
6
.
1
5
-
6
.
2
8
-
6
.
4
1
-
4
1
-
5
.
6
5
-
5
.
7
8
-
5
.
9
1
-
6
.
0
4
-
6
.
1
7
-
5
.
7
9
-
5
.
9
2
-
6
.
0
5
-
6
.
1
8
-
6
.
3
1
-
5
.
9
3
-
6
.
0
6
-
6
.
2
0
-
6
.
3
2
-
6
.
4
6
-
4
2
-
5
.
7
1
-
5
.
8
4
-
5
.
9
6
-
6
.
0
9
-
6
.
2
2
-
5
.
8
4
-
5
.
9
7
-
6
.
1
0
-
6
.
2
2
-
6
.
3
5
-
5
.
9
9
-
6
.
1
1
-
6
.
2
4
-
6
.
3
7
-
6
.
4
9
-
4
3
-
5
.
7
7
-
5
.
8
9
-
6
.
0
1
-
6
.
1
3
-
6
.
2
6
-
5
.
9
0
-
6
.
0
2
-
6
.
1
5
-
6
.
2
7
-
6
.
4
0
-
6
.
0
4
-
6
.
1
6
-
6
.
2
8
-
6
.
4
1
-
6
.
5
4
-
4
4
-
5
.
8
1
-
5
.
9
3
-
6
.
0
6
-
6
.
1
8
-
6
.
3
0
-
5
.
9
4
-
6
.
0
6
-
6
.
1
9
-
6
.
3
2
-
6
.
4
4
-
6
.
0
8
-
6
.
2
0
-
6
.
3
3
-
6
.
4
5
-
6
.
5
8
-
4
5
-
5
.
8
7
-
5
.
9
8
-
6
.
1
1
-
6
.
2
3
-
6
.
3
6
-
6
.
0
0
-
6
.
1
1
-
6
.
2
4
-
6
.
3
6
-
6
.
4
9
-
6
.
1
3
-
6
.
2
5
-
6
.
3
8
-
6
.
4
9
-
6
.
6
3
-
4
6
-
5
.
9
2
-
6
.
0
4
-
6
.
1
6
-
6
.
2
8
-
6
.
4
1
-
6
.
0
4
-
6
.
1
7
-
6
.
2
9
-
6
.
4
1
-
6
.
5
4
-
6
.
1
8
-
6
.
3
0
-
6
.
4
2
-
6
.
5
5
-
6
.
6
8
-
4
7
-
5
.
9
6
-
6
.
0
9
-
6
.
2
0
-
6
.
3
3
-
6
.
4
4
-
6
.
0
9
-
6
.
2
2
-
6
.
3
3
-
6
.
4
5
-
6
.
5
8
-
6
.
2
2
-
6
.
3
5
-
6
.
4
6
-
6
.
5
9
-
6
.
7
1
-
4
8
-
6
.
0
1
-
6
.
1
4
-
6
.
2
6
-
6
.
3
7
-
6
.
4
9
-
6
.
1
4
-
6
.
2
6
-
6
.
3
8
-
6
.
4
9
-
6
.
6
1
-
6
.
2
7
-
6
.
4
0
-
6
.
5
1
-
6
.
6
3
-
6
.
7
5
-
4
9
-
6
.
0
7
-
6
.
1
9
-
6
.
3
0
-
6
.
4
3
-
6
.
5
4
-
6
.
1
9
-
6
.
3
2
-
6
.
4
3
-
6
.
5
5
-
6
.
6
7
-
6
.
3
2
-
6
.
4
4
-
6
.
5
6
-
6
.
6
8
-
6
.
8
0
-
5
0
-
6
.
1
3
-
6
.
2
3
-
6
.
3
6
-
6
.
4
7
-
6
.
5
9
-
6
.
2
5
-
6
.
3
6
-
6
.
4
8
-
6
.
5
9
-
6
.
7
2
-
6
.
3
7
-
6
.
4
8
-
6
.
6
0
-
6
.
7
2
-
6
.
8
4
-
5
1
-
6
.
1
7
-
6
.
2
9
-
6
.
4
0
-
6
.
5
1
-
6
.
6
3
-
6
.
2
9
-
6
.
4
1
-
6
.
5
3
-
6
.
6
3
-
6
.
7
6
-
6
.
4
2
-
6
.
5
4
-
6
.
6
5
-
6
.
7
6
-
6
.
8
9
-
5
2
-
6
.
2
2
-
6
.
3
3
-
6
.
4
4
-
6
.
5
6
-
6
.
6
7
-
6
.
3
5
-
6
.
4
5
-
6
.
5
6
-
6
.
6
8
-
6
.
8
0
-
6
.
4
7
-
6
.
5
8
-
6
.
6
9
-
6
.
8
1
-
6
.
9
2
-
5
3
-
6
.
2
7
-
6
.
3
9
-
6
.
4
9
-
6
.
6
0
-
6
.
7
1
-
6
.
3
9
-
6
.
5
0
-
6
.
6
2
-
6
.
7
2
-
6
.
8
3
-
6
.
5
1
-
6
.
6
3
-
6
.
7
4
-
6
.
8
5
-
6
.
9
5
-
5
4
-
6
.
3
1
-
6
.
4
3
-
6
.
5
4
-
6
.
6
5
-
6
.
7
6
-
6
.
4
3
-
6
.
5
5
-
6
.
6
6
-
6
.
7
6
-
6
.
8
8
-
6
.
5
5
-
6
.
6
7
-
6
.
7
8
-
6
.
8
9
-
7
.
0
0
-
5
5
-
6
.
3
8
-
6
.
4
7
-
6
.
5
8
-
6
.
6
9
-
6
.
8
0
-
6
.
4
9
-
6
.
5
9
-
6
.
7
0
-
6
.
8
1
-
6
.
9
2
-
6
.
6
0
-
6
.
7
1
-
6
.
8
2
-
6
.
9
3
-
7
.
0
4
-
5
6
-
6
.
4
1
-
6
.
5
2
-
6
.
6
2
-
6
.
7
5
-
6
.
8
5
-
6
.
5
2
-
6
.
6
4
-
6
.
7
4
-
6
.
8
6
-
6
.
9
7
-
6
.
6
4
-
6
.
7
6
-
6
.
8
6
-
6
.
9
8
-
7
.
0
9
-
5
7
-
6
.
4
6
-
6
.
5
7
-
6
.
6
8
-
6
.
7
8
-
6
.
8
9
-
6
.
5
8
-
6
.
6
8
-
6
.
8
0
-
6
.
8
9
-
7
.
0
1
-
6
.
6
9
-
6
.
8
0
-
6
.
9
1
-
7
.
0
1
-
7
.
1
3
-
5
8
-
6
.
5
1
-
6
.
6
2
-
6
.
7
2
-
6
.
8
3
-
6
.
9
3
-
6
.
6
2
-
6
.
7
3
-
6
.
8
3
-
6
.
9
4
-
7
.
0
5
-
6
.
7
4
-
6
.
8
5
-
6
.
9
5
-
7
.
0
6
-
7
.
1
6
-
5
9
-
6
.
5
6
-
6
.
6
6
-
6
.
7
7
-
6
.
8
7
-
6
.
9
8
-
6
.
6
7
-
6
.
7
7
-
6
.
8
9
-
6
.
9
8
-
7
.
0
9
-
6
.
7
8
-
6
.
8
8
-
7
.
0
0
-
7
.
1
0
-
7
.
2
1
-
6
0
-
6
.
6
0
-
6
.
7
1
-
6
.
8
2
-
6
.
9
2
-
7
.
0
2
-
6
.
7
1
-
6
.
8
2
-
6
.
9
3
-
7
.
0
3
-
7
.
1
3
-
6
.
8
2
-
6
.
9
3
-
7
.
0
4
-
7
.
1
5
-
7
.
2
5
23
0
−5
−10
−15
−20
−25
−30
0
−5
−10
−15
−20
−25
−30
−5.5
−5
−4.5
−4
−3.5
−3
c
1
c
2
C
r
i
t
i
c
a
l

V
a
l
u
e
c
2
c
1

5

4
.
8

4
.
6

4
.
4

4
.
2

4

3
.
8

3
.
6

3
.
4
0 −5 −10 −15 −20 −25 −30
−5
−10
−15
−20
−25
−30
Figure 1: Critical values at the five percent level for the AEG test as a function of c
1
and c
2
. The
top panel shows the surface describing the five percent critical values of the AEG test, in the case
of an intercept and one regressor, when c
1
and c
2
are non-identical. The bottom panel shows the
corresponding contour plot. The values are based on 10, 000 repetitions with T = 1, 000.
0 −5 −10 −15 −20 −25 −30
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S
i
z
e
c


T=100, n=1 T=100, n=3 T=500, n=1 T=500, n=3
Figure 2: Size properties of the Engle and Granger (1987) test of cointegration, as a function of the
local-to-unity parameter c. The graph shows the average rejection rates under the null hypothesis of
no cointegration for the Engle and Granger test of cointegration, i.e. the standard AEG test evaluated
under the assumption that c = 0, for different true values of c. The sample size is equal to either
T = 100 or 500, and the number of regressors equal to either n = 1 or 3. The true persistence in the
data is equal to C = diag (c, ..., c), where c varies between 0 and −30. The results are based on 10, 000
repetitions.
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
T=100, n=1
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
T=100, n=3
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
T=500, n=1
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
T=500, n=3
S
i
z
e
c


α
1
= 0.75 α
1
= 0.50 α
1
= 0.25
Figure 3: Size properties of the Bonferroni AEG test when the variables all have equal persistence.
The graphs show the average rejection rates for the Bonferroni AEG test, under the null hypothesis
of no cointegration, for α
1
= 0.75, 0.50, and 0.25. The sample size is equal to either T = 100 or 500,
and the number of regressors is equal to either n = 1 or 3. The true persistence in the data is equal
to C = diag (c, ..., c), where c varies between 0 and −30. The results are based on 10, 000 repetitions.
−20 −15 −10 −5 0 5 10
0
0.05
0.1
0.15
c=−5
c
−25 −20 −15 −10 −5 0 5
0
0.02
0.04
0.06
0.08
0.1
0.12
c=−10
c
−50 −40 −30 −20 −10 0
0
0.02
0.04
0.06
0.08
c=−20
c
−60 −50 −40 −30 −20 −10 0
0
0.01
0.02
0.03
0.04
0.05
0.06
c=−30
c


75 Percent 50 Percent 25 Percent
Figure 4: Estimates of the lower bounds of c. The graphs show the density of the estimates of the
lower bounds of c, with confidence levels of 75, 50, and 25 percent, based on inversion of the DF-GLS
statistic. The results are obtained from 10, 000 simulations of a univariate local-to-unity process, with
local-to-unity parameter c, iid normal innovations and sample size T = 500.
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
0.12
T=100, n=2
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
0.12
T=100, n=2
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
0.12
T=500, n=3
S
i
z
e
c
0 −5 −10 −15 −20 −25 −30
0
0.02
0.04
0.06
0.08
0.1
0.12
T=500, n=3
S
i
z
e
c


α
1
= 0.75 α
1
= 0.50 α
1
= 0.25
Figure 5: Size properties of the Bonferroni AEG test when c
i
is not identical for all i. The graphs show
the average rejection rates for the Bonferroni AEG test, under the null hypothesis of no cointegration,
for α
1
= 0.75, 0.50, and 0.25. The sample size is equal to either T = 100 or 500, and the number
of regressors is equal to either n = 2 or 3. For n = 2, the true persistence in the data is equal to
C = diag (c
1
, −10, −20), and for n = 3, C = diag (c
1
, 0, −10, −20) , where c
1
varies between 0 and
−30. The results are based on 10, 000 repetitions.
0.5 0.6 0.7 0.8 0.9 1
0
0.25
0.5
0.75
1
Equal Persistence: T=100, n=1
P
o
w
e
r
ρ
0.8 0.9 1
0
0.25
0.5
0.75
1
Equal Persistence: T=500, n=1
P
o
w
e
r
ρ


c=−2
c=−10
c=−20
0.5 0.6 0.7 0.8 0.9 1
0
0.25
0.5
0.75
1
Different Persistence: T=100, n=3
P
o
w
e
r
ρ
0.8 0.9 1
0
0.25
0.5
0.75
1
Different Persistence: T=500, n=3
P
o
w
e
r
ρ
Figure 6: Power properties of the Bonferroni AEG test. The graphs show the average rejection rates
of the Bonferroni AEG test, for α
1
= 0.50, under the alternative of cointegration. The power is plotted
as a function of ρ, the AR(1) persistence parameter in the cointegrating residuals. The sample size
is set equal to either T = 100 or 500. The left column gives results for the case of one regressor with
persistence C
2
= −2, −10, or −20. The right column gives the results for the case with three regressors
and C
2
= diag (0, −10, −20). The results are based on 10, 000 repetitions.